Hi,

In time series analysis before modelling the series the following needs to be done.

1. To examine whether seasonality is present. There are several techniques available for this; such as seasonality ratio, seasonality indicator, Gini coefficient and the seasonality index. I prefer seasonality index over others. If the seasonality is present, then you need to address this in your time series model.

2. Stationarity tests- You can use ADF, DF-GLS, KPSS and Phillips-Perros test. If the variables are not stationarity you need to take the 1st difference. Normally in the 1st difference most of the series will be stationary. You can try various transformations of the series such as in levels, logs, log difference and annual-log difference to see which transformation looks better for your analysis.

3. Depending on your research questions there are many techniques available. If you state your research question I can further provide information about such techniques.

Kind regards

Thushara

Hi,

As has been mentioned in the previous answer, the exact nature of your analysis depends on your research objectives. Nevertheless, the following characteristics of time series are analysed as preliminary testing: 

Stationarity, Seasonality, Trend analysis, etc. Autocorrelation function ACF can be used to test both stationarity & seasonality in a time series.

first of all see the graph of your time series often it reveals a lot

The time plot is invariably the first thing to do in time series analysis. It reveals a lot about the series. Salient features of a time series are features that matter most about the series. They are the basic features about the series. They are features that give indication of how to model the series. The traditional components: trend, seasonality and cyclical movement are, to me, such salient feature. The traditional approach to time series analysis involves identifying and separately estimating these components of the series. There are two basic models to adopt: the additive model or the multiplicative model. Trend may be measured by least squares approach or the moving average method. The series can be de-trended with a view to computing the seasonal index. This traditional approach is the initial way to study the salient features of a time series.